8 APPLICATION: THE NUMBER e 171
7. Let I = [0,1 [ and let E be the vector space of regulated complex functions defined in I,
bounded and continuous on the right (i.e. /(*+) =/(0 for t E I).

JH-l -------
f(t)g(t) dt is a nondegenerate positive hermitian form
-i

(see (8.5.3)). Prove that the prehilbert space E thus defined is not complete (use the fact
that the function equal to sin(l/r) for / > 0, to 0 for t = 0, is not in E).
(b) Define the sequence (fn) of elements of E in the following way:

(1) /o is the constant 1;
(2) for each integer n > 0, let m be the largest integer such that 2m ^ n, and let
2k 2k 4- 1
n = 2m + k; fn is taken as equal to 2m/2 for —T < f <-----TT-, to —2m/2 for
2m 2m
2k + 1 2k + 2
2»»+i 2m4
Prove that in the prehilbert space E, (fn) is an orthonormal system (the **Haar
orthonormal system ").

(c) For each n ^ 0, let Vn be the supspace of E generated by the fk of indices k ^ n.
Show that there is a decomposition of I into n +1 intervals of type [a, /3[ without
common points, such that, in each of these intervals, every function belonging to Vn is
constant; conversely, every function having that property belongs to Vn (consider the
dimension of the vector subspace of E generated by these functions).

(d) Let g be an arbitrary function of E, h its orthogonal projection (Section 6.3) on Vn;
show that in each of the intervals [a, j3[ in which all the functions of Vn are constant,

1
"*'£-«
(e) Show, by using (d), that for any function g e E which is continuous in I, the series
of general term (g \ fyfn(t) is uniformly convergent in I and that its sum is equal 1
Conclude from that result that (fn) is a total orthonormal system in E.

8. Let/be a regulated real valued function in a compact interval I = [a, b]; let \f(t)\ dt =
Ja
c. Show that for any e > 0, there is a real valued continuous function g in I, such that
ffr
\g(t)\ ^ 1 in I, and that f(t)g(t) dt ^ c e. (Reduce the problem to the case in which/

Ja
is a step function.)
8. APPLICATION: THE NUMBER e
For any number a > 0, the function x ->• a* is continuous in R (Section
4.3), hence the function g(x) = f*a* dt is defined and differentiable in R, with

/•jc+l px rx+l
g'(x) = ax everywhere. Now we have g(x + 1) = cfdt—\aldt + \ a1 dt.
JO JO J x
But by (8.7.4), f^a' dt = ^ ax+u du = ax ^ a" du; as a*^inf(a, 1) for
Jx Jo JO
0 < x ^ 1, c = f a14 du is > 0 by (8.5.3), hence we can write
Jo